Fault diagnosis method for incremental faults in liquid rocket engine turbopumps

Abstract

As the complexity of modern liquid rocket engines continues to increase, equipment faults are becoming increasingly diverse, while their occurrence stages and underlying causes are more difficult to predict. This poses significant challenges for the fault diagnosis of critical components such as turbopumps. Traditional fault diagnosis methods struggle to address these challenges, particularly in handling newly emerging fault types during long-term operation. To tackle this issue, this paper proposes a Regionally Variable Continual Learning Classification Algorithm (RVCLCA) based on an artificial immune system. The proposed method integrates an improved Transformer with the DBSCAN clustering algorithm, enabling effective identification and diagnosis of turbopump faults in liquid rocket engines, with notable advantages in recognizing novel fault types. First, the continual classification performance of RVCLCA is validated on well-known UCI datasets, where the results demonstrate that RVCLCA outperforms conventional classification methods, achieving a maximum continual learning accuracy of 93.33%. Finally, the effectiveness of the proposed algorithm is further verified through turbopump fault diagnosis experiments, highlighting its potential for practical applications in liquid rocket engines.